GCC Lands Cannonlake, Skylake Costs; LLVM/Clang Gets Intel CET

In addition to the GCC plugin support on Windows/MinGW, there are more compiler happenings this weekend.

Hitting mainline GCC since that earlier post about the MinGW plugin support is this commit landing the -march=cannonlake target for these next-gen Intel CPUs. It's among the many GCC 8 features and previously covered the Cannonlake enablement while now it's been merged to mainline.

Over Skylake/Kabylake, notable to Cannonlake is now AVX-512 support being widely available rather than just limited to the server/X-Series CPUs. It's great to see AVX-512 support rolling out to Intel's entire desktop line-up. Cannonlake also adds the SHA extensions, CLWB persistent memory instructions, and UMIP for user-mode instruction prevention.

This Cannonlake commit also adds Skylake cost tables for generating better tuned code for Skylake CPUs and newer.

Meanwhile, on the LLVM compiler side, Intel landed their work on CET support in LLVM. CET is the Control-flow Enforcement Technology that already landed in GCC with this security feature aimed to prevent return-oriented programming and call-jump-oriented programming attacks.

CET will be supported in future Intel CPUs and now the core LLVM bits have landed as well as the Clang work. That work will be found in LLVM/Clang 6.0 due out early next year followed by GCC 8.1 one or two months later.